Most organizations want employees using AI—and right now—but many want to better understand security risks and environmental impact of AI-first workloads. A critical gap becomes apparent: the need to navigate the complex trade-offs between computational power, data security, and resource consumption. aihorizon R&D, an independent research and development organization based in Freiburg and Munich, Germany, is pioneering frameworks with research into responsible AI and developing solutions for it.

They’re asking: How do we balance high-utility AI design with the technical requirements of data privacy and resource efficiency?

For Dr. Michael Jülich, Founder & Chief Scientist at aihorizon R&D, responsibility is not a secondary constraint; it is the core architectural blueprint.

Tackling responsible AI demands a rigorous rethinking of how tech is deployed. The exploration spans everything from powerful cloud-based large language models (LLMs) to the emerging generation of AI PC chips.

Security and resource efficiency are two of the biggest challenges in AI today. While cloud-based LLMs offer immense power, they also bring growing concerns regarding data sovereignty and the significant environmental footprint of large-scale data centers.

These challenges sparked the interest of Dr. Jülich, who was then the Director of Digital Transformation at Educational Foundation Freiburg. Dr. Jülich and Volker Leitzgen, CTO of Microsoft Tech for Social Impact, collaborated on a project to train students in ethical AI use. Their efforts gave rise to the creation of aihorizon R&D, now led by Dr. Jülich.

aihorizon R&D has since broadened its focus beyond education to help industry and NGOs address two critical questions:

• Data sovereignty: Can sensitive information be processed entirely on-device, ensuring it never leaves the user’s control?
• Environmental transparency: Can AI’s CO2 impact be monitored and optimized in real time without compromising performance?

To address these issues, aihorizon R&D launched TerraCore, a collaborative project with input from leading industry partners like Microsoft and Educational Foundation Freiburg, with support from Qualcomm Technologies, Inc.

The TerraCore project addresses the critical imperatives of modern AI, providing users with unprecedented transparency into both their data privacy and their energy consumption. The Responsible AI (rAI) app serves as a command interface, empowering users to actively negotiate the privacy and environmental parameters of their AI engagement by enabling them to:

• Choose whether to run an AI task in the cloud or on the device
• Instantly evaluate the privacy trade-offs and real-time carbon emissions for each task
• Get a clear understanding of where every prompt goes and its associated carbon footprint

This app enables organizations to define guardrails that foster awareness and responsible usage. As the aihorizon R&D team puts it:

“CO2 tracking and data privacy are not just technical features—they are systemically embedded into the user experience.”

TerraCore helps transform responsible AI from theory to possibility through a lived, practical tool.

PCs powered by Snapdragon X Series chips are making on-device AI possible for the TerraCore initiative. Qualcomm Technologies designed its Snapdragon compute platform specifically to run AI locally in a closed, private hardware environment. Now available in a variety of Windows PCs, Snapdragon compute chips come with a built-in NPU (a neural processing unit optimized for AI workloads) that delivers up to 80 TOPS (trillion operations per second), beyond the standard that Microsoft sets for Copilot+ PCs. That is the key to running an LLMbased AI right on the device instead of from the cloud with the needed acceleration and efficiency.

With these AI PCs, aihorizon R&D was able to fully use their rAI app, a software tool they designed that lets users run the same AI workflow locally or in the cloud, then observe and measure differences.

For example, within the rAI app on PCs powered by Snapdragon processors, users could give the same prompt to a cloudbased LLM and to an LLM running natively, then see the difference in energy consumption, security exposure, and performance.

What becomes possible: Local LLM inference runs directly on the PC

What this means: Instant feedback loops with zero latency from remote services

What becomes possible: Proprietary and sensitive information stays on the device

What this means: Data remains under user control

What becomes possible: Carbon footprint becomes transparent and actionable

What this means: Users can make informed choices

This fusion of the right hardware and bold vision is enabling users to understand the impacts of AI and provides them with alternatives.

In the ongoing responsible AI workshops at Educational Foundation Freiburg, teachers and students use the rAI app to see the implications of their AI choices. Via the rAI app, they can select cloud vs. local AI and see the impact on:

• Environmental metrics: See if and how opting for local AI can save energy and CO₂ emissions.
• Privacy module: Instantly detect sensitive data and make suggestions on what to keep local.
• Performance clarity: Clearly see when local or cloud is the right solution.

The team’s mission is clear: to make the responsible use of AI visible, understandable, and actionable for everyone. For the first time, schools can securely leverage AI to handle sensitive items like personal and administrative data.

Students have discovered a new level of agency—making deliberate choices about privacy and energy, not just accepting defaults. They can now generate prompts using both a cloud-based LLM and an on-device LLM, allowing them to choose options that consume less energy.